Why do Analysts Issue Long-term Earnings Growth Forecasts?

An Empirical Analysis

Douglas O. Cooka, Huabing (Barbara) Wang

b

a University of Alabama, Tuscaloosa, AL 35487-0224, USA b University of Alabama, Tuscaloosa, AL 35487-0224, USA

ABSTRACT

We examine analysts’ motives to issue long-term earning growth (LTG) forecasts. We

find that analysts are more likely to issue LTG forecasts when their incentive to please

managers is strong. In addition, analysts are more likely to choose firms that they are

more optimistic about for LTG coverage. We find mixed evidence regarding whether

analysts issue LTG forecasts to signal their ability or to meet investors’ informational

needs. Augmenting Ljungqvist et al (2006), we show that LTG forecasts are issued less

likely to please managers, but more likely to meet investors’ information needs in the

presence of high institutional ownership.

2

1. Introduction

While the extant literature (e.g., Chan, Karceski and Lakonishok (2003)) yields

overwhelming evidence on the over-optimism and inaccuracy of long-term earnings

growth (LTG) forecasts, it remains silent on why analysts issue these forecasts, a question

that becomes even more intriguing given the more voluntary nature of LTG forecasts

compared with their near-term counterparts. That is, why do some analysts issue for some

companies LTG forecasts, which are often deemed as extremely inaccurate and overly

optimistic, when they can choose not to? This study offers insights into this question by

empirically examining four non-exclusive hypotheses: analysts issue LTG forecasts to

signal their ability, to reveal their optimism, to please the management (since these

forecasts are overly optimistic), and to satisfy investors’ informational needs.

With one-year-ahead annual earnings forecasts as the benchmark sample, we test

our hypotheses jointly in a fixed-effect framework with analyst-year (or analyst) effect

fixed to ensure that our results are not driven by unobserved analyst-level heterogeneity

such as analyst peculiarities.

We document evidence for the manager pleasing and optimism revealing hypothesis,

but mixed results for the analyst ability signaling and investor informational needs

satisfying motives. Augmenting Ljungqvist et al (2006)’s finding about institutional

investors’ moderating role in analyst research, we find that analysts are less (more) likely

to issue long-term forecasts for companies with large institutional ownership to please

managers (to meet investors’ information needs).

3

Our paper contributes to the literature in several ways. First, our results suggest

that LTG forecasts may serve as a manipulative tool for analysts to please managers.

Therefore, conflicts of interest may affect not only the quality of analyst research, such as

the biases of analyst recommendations as examined by previous literature, but also the

type of information included in the analyst reports. This motive may partly explain the

documented over-optimism in LTG forecasts.

An examination of the providence of LTG forecasts offers several advantages in

the investigation of interest conflicts. For example, due to reputation concerns, analysts

are less likely to bias their near-term forecasts or recommendations. However, with

accuracy, and thus reputation loss, not a primary concern, the voluntarily provided LTG

forecasts provide a cleaner setting to study motives related to conflict of interest.

Furthermore, the quality of analyst earnings forecasts and recommendations may depend

not only on analyst incentives but also on analyst ability and even factors beyond

analysts’ control. For example, less able or less fortunate analysts may appear to issue

biased recommendation in absence of incentives to please managers. The decision to

provide LTG forecasts, however, is not affected by so many complicating influences.

Instead, it is totally in analysts’ control and involves little analyst ability.

Furthermore, our results augment Ljungqvist et al (2006)’s finding about the role of

institutional investors in analyst research. We find evidence that higher institutional

ownership reduces the likelihood of analysts issuing LTG forecasts to please mangers.

Furthermore, we show that the presence of higher institutional ownership makes analysts

4

more responsive to investors’ information needs.

The remainder of the paper proceeds as follows. Section 2 develops hypotheses.

Section 3 discusses our data, sample, variables, and summary statistics. Section 4

presents the main results. Section 5 examines the role of institutional investors in

analysts’ motives of LTG forecast issuance. Section 6 concludes.

2. Hypotheses development

2.1 Characteristics of LTG forecasts

There is a growing body of literature on LTG forecasts. La Porta (1996) finds that

investment strategies seeking to exploit errors in analysts' forecasts earn superior returns

because expectations about future growth in earnings are too extreme. Dechow and Sloan

(1997) also document that naive reliance on analysts' forecasts of future earnings growth

can explain over half of the higher returns to contrarian investment strategies. Harris

(1999) reports three characteristics of LTG forecasts: (1) they are extremely low in

accuracy; (2) they are inferior to the forecasts of a naïve model in which earnings are

assumed to follow a martingale, and (3) they are significantly over-optimistic, exceeding

the actual growth rate by an average of seven percent per annum. Chan, Karceski and

Lakonishok (2003) analyze historical long-term growth rates across a broad cross section

of stocks and show that I/B/E/S growth forecasts are overly optimistic and add little

predictive power.

5

In the setting of IPOs, prior literature suggests that conflict of interests plays an

important role in the optimism of LTG forecasts. For example, Rajan and Servaes (1997)

examine data on analyst following for a sample of initial public offerings completed

between 1975 and 1987, and find that analysts are overoptimistic about the earnings

potential and long-term growth prospects of recent IPOs. They further document that, in

the long run, IPOs have better stock performance when analysts ascribe low growth

potential rather than high growth potential. Lin and McNichols (1998) find that lead and

co-underwriter analysts' growth forecasts and recommendations are significantly more

favorable than those made by unaffiliated analysts, although their earnings forecasts are

not generally greater. Purnanandam and Swaminathan (2004) also document that, ex post,

the projected high growth of overvalued IPOs fails to materialize, while their profitability

declines from pre-IPO levels. Their results suggest that IPO investors are deceived by

optimistic growth forecasts and pay insufficient attention to profitability in valuing IPOs.

2.2 Why do analysts issue LTG forecasts?

In this section, we develop four non-exclusive testable hypotheses about the supply

of long-term forecasts, which are analyst ability signaling, optimism revealing,

management pleasing, and investor information needs satisfying. We also discuss the role

of analyst peculiarity in LTG forecast issuance.

6

A) Analyst ability signaling

At first sight, it may seem reasonable that the highly inaccurate and optimistic LTG

forecasts are associated with low-quality analysts. However, while LTG forecasts are

highly inaccurate and overly optimistic ex post, they may provide useful information to

investors when they are published. The huge errors we observe ex post might just reflect

the difficulty in projecting earnings growth far into the future.

Besides, analysts don’t have to provide LTG forecasts. Since it is a challenging job

to forecast the far future, only high-ability analysts are confident enough to issue LTG

forecasts. Therefore, we argue that analysts are more likely to issue LTG forecasts when

they are of higher ability, or at least, they perceive themselves as of higher ability.

H1: Analysts of higher ability are more likely to issue LTG forecasts.

B) Analyst optimism revealing

McNichols and O'Brien (1997) find evidence of self-selection bias in analyst

coverage. Specifically, they show that analysts tend to add firms they view favorably and

drop firms they view unfavorably. Along the same line of thinking, we argue that there is

a self-selection bias in the providing of LTG forecasts as well. After all, analysts should

have stronger incentives to collect long-term company-specific information when they

are confident in the company’s future.

The documented optimistic nature of LTG forecasts also appears to suggest that

analysts who are more optimistic about the company are more likely to issue long-term

forecasts. Thus, we expect analysts to be more likely to issue LTG forecasts when they

7

are more optimistic about the company’s future.

H2: Analysts are more likely to issue LTG forecasts for companies they are more

optimistic about.

C) Management Pleasing

In practice, sell-side analysts often find themselves serving two masters. On the one

hand, they serve investors, and thus aim at providing accurate and reliable research. On

the other hand, their incentives to please the managers often obscure their goal of

“objectivity”, making the company they cover their other master. At the very least,

analysts are often afraid to offend managers by providing unfavorable opinions partially

because managers may withhold information from those analysts they are unhappy with

(e.g., Lim (2001)).

In addition to informational concerns, analysts face an even higher stake when the

company they cover is also an investment banking customer of the investment bank the

analysts are affiliated with. There is a growing body of literature examining the role

interest conflict plays in various aspects of analyst research. Dugar and Nathan (1995)

show that analysts whose employers have an investment banking relationship with a

company issue more favorable recommendations. Lin and McNichols (1998) find that

lead and co-underwriter analysts' growth forecasts and recommendations are significantly

more favorable than those made by unaffiliated analysts, although their earnings forecasts

are not generally greater. Michaely and Womack (1999) document that stocks that

underwriter analysts recommend perform more poorly than 'buy' recommendations by

8

unaffiliated brokers prior to, at the time of, and subsequent to the recommendation date,

and further show that the market does not recognize the full extent of this bias. Agrawal

and Chen (2005a) find that potential investment banking relationship has no effect on

quarterly earnings forecasts, but is positively associated with more optimistic long-term

growth forecasts. Agrawal and Chen (2005b) show that analyst recommendation levels

are positively associated with the magnitude of conflicts they face, but investors

recognize analysts’ conflicts and properly discount analysts’ opinions. O'Brien,

McNichols and Lin (2005) find that affiliated analysts are slower to downgrade from the

“Buy” and “Hold” recommendations and significantly faster to upgrade from the “Hold”

recommendations. James and Karceski (2006) document that underwriter-affiliated

analysts provide protection in the form of "booster shots" of stronger coverage if the IPO

firm experiences poor aftermarket stock performance. Ljungqvist et al (2006) confirm the

positive relation between investment banking and brokerage pressure and analyst

recommendations, and further show that both bank reputation and institutional investors

serve as moderating forces that temper analyst optimism.

Regarding LTG forecasts, prior literature also finds substantial evidence that

investment banking relationship contributes to the extreme optimism in long-term

earnings growth forecasts (e.g., Rajan and Servaes (1997) and Purnanandam and

Swaminathan (2004)). Agrawal and Chen (2005a) suggest that analysts do not respond to

conflicts by biasing short-term (quarterly EPS) forecasts, but appear to succumb to

conflicts when making LTG forecasts. After all, in the case of LTG forecasts, which are

9

often neglected by investors who put heavy weight on analyst near-term forecasts and

recommendations, there is only one master left: the company they cover. Furthermore,

given that LTG forecast are relatively difficult to verify ex post, the reputation loss

associated with an inaccurate LTG forecast is minimal.

One may argue that analysts should be indifferent to LTG forecast issuance because

these forecasts are generally ignored by investors and thus do not benefit managers at the

cost of investors. However, conflict of interest, although behavior-altering, does not

necessarily affect the interest of the third party. Instead, it is rational for analysts to

respond to conflict of interest in a way less harmful to investors. The voting behavior of

mutual fund managers documented by Davis and Kim (2006) may lend support to this

view. Specifically, Davis and Kim (2006) find that mutual fund managers appear to side

with management especially when there is no clear evidence that the measure being voted

on have an impact on shareholder wealth. Therefore, we argue that, due to the general

ignorance by investors, LTG forecasts may be subject to analyst manipulation to please

the companies they cover.

H3: The supply of (optimistic) LTG forecasts is positively related to analysts’

incentive to please managers.

D) Investor Information Need Satisfying

Defond and Hung (2003) document that financial analysts respond to market-based

incentives to provide investors with value-relevant information. In particular, they find

that analysts tend to forecast cash flows for firms whose accounting, operating and

10

financing characteristics suggest that cash flows are useful in interpreting earnings and

assessing firm viability. Along the same line, we expect that analysts provide LTG

forecasts for firms whose long-term prospects are especially important for the valuation

of their stocks. Therefore, we expect companies with large growth options to be more

likely to receive LTG forecasts.

H4.1: Companies with larger growth options are more likely to receive LTG

forecasts.

Meanwhile, Ljungqvist et al (2006) suggest that institutional investors serve as the

ultimate arbiters of an analyst’s reputation. Furthermore, institutional investors tend to be

sophisticated users of the information analysts provide, who are therefore more likely to

demand long-term information in their decision process. Consequently, analysts should

be more likely to supply detailed research including a firm’s long-term prospects when

they know that the report is more likely to be read by institutional investors. Therefore,

we expect companies with higher institutional investor ownership to be more likely to

receive LTG forecasts.

H4.2: Companies with higher institutional investor ownership are more likely to

receive LTG forecasts.

E) Analyst peculiarity

In addition to the four hypotheses we develop above, it is possible that the issuance

of LTG forecasts depends on the peculiarities of analysts, such as their working habits

and tastes. If this is true, we should find no systematic pattern in the issuance of LTG

11

forecasts. In addition, we should find little variation in the issuance decision of a

particular analyst covering several companies.

2.3 Institutional investors’ role in analysts’ motives to issue LTG forecasts

Ljungqvist et al (2006) document the role of institutional investors in moderating

conflicts of interest in analyst research. They argue that driven by their career concerns,

analysts are less likely to succumb to investment banking pressure in stocks that are

highly visible to their institutional investor constituency.

In addition, underlying our hypotheses, we assume that long-term forecasts can be

manipulated because the little attention they receive from investors. However, unlike

individual investors, who may be more focused on analyst recommendations and

near-term earnings forecasts while totally neglecting long-term forecasts, institutional

investors read analyst reports thoroughly and put more weights on the contents instead.

Consistently, Mikhail, Walther, and Willis (2006) find evidence that large investors are

more sophisticated processors of information, while small investors are more easily

misled by analyst research. Therefore, we expect analysts less likely to issue LTG

forecasts to please managers for companies heavily owned by institutional investors. For

the same reason, we also expect the presence of institutional investors to enhance

analysts’ incentives to issue LTG forecasts when long-term information is valuable to

investors.

Overall, we hypothesize that the presence of institutional investors is negatively

12

(positively) relate to analysts’ manager-pleasing (investor information needs satisfying)

motives to issue LTG forecasts.

H5: Analysts are less (more) likely to issue LTG forecasts to companies with large

institutional ownership to please managers (to meet investors’ information needs).

3. Data, sample, variables, and summary statistics

3.1 Data and sample

As in Defond and Hung (2003), we collect one-year-ahead annual earnings

forecasts (FY1) as our benchmark sample to control for other factors that affect the

availability of LTG forecasts.1 We collect the one-year-ahead annual earnings forecasts

in the I/B/E/S detail history file from year 1991 to 2003. We identify each

analyst-firm-(forecast) year combination2 and check whether there is any LTG forecast

associated with these analyst-firm-year combinations. LTG forecasts are the long-term

earnings growth forecasts as collected by I/B/E/S, which usually covers a five-year

period that begins on the first day of the current fiscal year.

Panel A of Table 1 reports the number and proportion of firm-analyst pairs, analysts,

1 The LTG forecasts, as collected by I/B/E/S, usually cover a five-year period that begins on the first day

of the current fiscal year.

2 Instead of using the year for which a forecast is made, we use the year during which a forecast is made.

For example, the time stamp for a one-year-ahead forecast that is made in 2000 but for the Dec. 2001 fiscal

quarter will be 2000 instead of 2001. We do so because we expect the decision to supply the forecasts are

more economically related to the factors prevalent during the time the estimations are made

13

and firms associated with LTG forecasts by year. We observe significant variations in the

size of the benchmark sample over the sample period. However, the proportions of

analyst-firm associated with LTG forecasts demonstrate only small variations over years

except for year 2003, which is associated with the lowest proportion of LTG forecast

coverage. Specifically, the proportion of firm-analyst pairs that are associated with LTG

forecasts is in the 42-47 percent range over period 1991-2002. Analysts who issue LTG

forecasts account for around 58 percent of all the analysts who issue one-year-ahead

earnings forecasts each year. The number of firms receiving analyst one-year-ahead

forecasts peaked in 1996 with 1,149 firms covered, but dropped dramatically thereafter.

In 2003, only 280 firms receive one-year-ahead forecasts from any analysts. The

proportion of firms receiving LTG forecasts also seems to decrease over time.

3.2 Variables

(a) LTG Issuance

LTG is a dummy variable that equals one if the observation is associated with

long-term earnings growth forecasts (LTG) as reported in I/B/E/S, and zero otherwise.

(b) Analyst Ability

We adopt three sets of analyst ability measures. The first is analyst experience,

which is adopted by many prior studies as proxies for analyst ability and skill. For

example, Clement (1999) finds that forecast accuracy is positively associated with

analysts' experience. Mikhail, Walther and Willis (2003) find that analysts underreact to

14

prior earnings information less as their experience increases, suggesting one reason why

analysts become more accurate with experience. Following prior literature, we introduce

two experience measures. The general experience of the analysts (Exp1) is defined as the

number of years the analysts have issued earnings forecasts of any type for any company

since 1983, when the sample period of I/B/E/S starts. Analysts’ firm-specific experience

(Exp2) equals the number of years the analysts have issued earnings forecasts of any type

for the company since 1983.

Second, we use the accuracy of the analyst’s previous near-term forecasts as a proxy

for analyst ability. Prior studies generally suggest persistence in analysts’ stock picking

and earnings forecasting ability. For example, Sinha, Brown and Das (1997) document

persistence in earnings forecast accuracy, that is, superior earnings forecasters in one

period tends to be superior the next period. Mikhail, Walther and Willis (2004) find that

analysts whose recommendation revisions earned the most (least) excess returns in the

past continue to outperform (underperform) in the future. Therefore, we adopt the

accuracy of the analysts’ past near term earnings forecasts for the same company to proxy

for analyst quality. We define net forecast error (NFE) as 100 times the absolute value of

the difference between the actual earnings and the analyst forecasts divided by the

company’s stock price the company’s stock price at the end of the previous fiscal year.

Past_NFE equals NFE 1t−

, that is, the net forecast error of the most recent near-term

15

earnings forecasts made during the previous year.3 We expect a positive (negative)

relation between the experience variables (Past_NFE) with the likelihood of long-term

forecast issuance.

Finally, analysts affiliated with prestigious brokers tend to be of higher quality, as

suggested by prior studies (e.g., Clement (1999)). We use the analysts’ brokerage house

affiliation as the other proxy for analyst ability. We collect the broker names that appear

as top 15 in “the leader list” of the Institutional Investor magazine (II) from year 1990 to

year 2002. If a broker appears as top 15 on “the leader list” of Institutional Investor in

year t, the broker is defined as high status broker for year t+1. The dummy variable Top15

takes on value one for analysts affiliated with the high status brokers and zero otherwise.

(c) Analyst Optimism

We adopt the optimism in analysts’ near-term forecasts to measure analyst optimism

about the company. Given the management’s incentive to manage market expectations

and to beat analyst forecasts, analysts who are optimistic to please managers should be

forced to restrict or even discontinue their optimism in near-term forecasts, and therefore,

we argue that the optimism in near-term forecasts should mostly capture the analysts’

genuine optimism. Specifically, we use the forecast bias the analysts reveal in their past

near-term forecasts to measure the analysts’ optimism towards the company. Forecast

Bias (FB) is 100 times the difference between the actual earnings and the analyst

3 When we use the average NFE over the three-year period prior to the year under consideration as an

alternative measure, the sample size is reduced, but the main results remain largely unchanged .

16

forecasts divided by the company’s stock price at the end of the previous fiscal year. A

negative (positive) FB indicates that the forecast overestimate (underestimate) the actual

earnings, and that it is optimistic (pessimistic). We define FB 1−t as the past near-term

forecast accuracy (Past_FB).4 We expect the estimated coefficient to be negative. That is,

increased analyst optimism, as measured by a more negative value of forecast bias, is

associated with higher likelihood of long-term forecast issuance.

(d) Management Pleasing Incentives

We adopt the existence of equity underwriting relationship as a proxy for analysts’

incentive to please the managers, and hypothesize that analysts are more likely to issue

long-term forecasts for firms who are also their investment banking customers.

We extract all the new common stock issues in the U.S. market from 1989 to 2004

from the Securities Data Company (SDC) new issues database. We hand match the

underwriters in the SDC database with the brokers in the I/B/E/S database. To enhance

the quality of our match, we obtain the starting and ending dates of the appearance of the

underwriter in the SDC database, and compare them with the starting and ending dates of

the appearance of the broker in the IBES database. We also check the merger and

acquisition history of the investment banks from the investment bank’s website as well as

by Google searching.5 We are able to get a one-to-one match for most of the SDC

4 When we use the average FB over the three-year period prior to the year under consideration as an

alternative measure, the sample size is reduced, but the main results remain largely unchanged 5 We also double check the matching with the investment bank M&A and name changes data complied by

Cheolwoo Lee, who generously provides us with the data.

17

underwriters. For underwriters/brokers that have experienced mergers or acquisitions, we

assume that the surviving investment banks/brokers inherit the investment banking

business and research coverage from both the acquirer and the target to assure continuity

if the target broker coverage stops at the year of the merger.

We assume that there is an investment banking relationship between the broker and

the firm from one year before the issuing of the new common stock to one year after. We

define IB as a dummy variable that equals one if the analyst is affiliated with the

investment bank that serves as a book runner for the company’s new common stock

issues, and zero otherwise. Considering that it is possible for analysts to issue LTG

forecasts for IPO firms because investors are in greater needs for long-term information

of these companies, we introduce an IPO dummy. Specifically, IPO equals one for

company i in year t if the company has an initial public offering as indicated by the IPO

flag in SDC for year t and t-1, and zero otherwise.

(e) Firm Growth Options

We adopt a firm’s capital expenditure and R&D expenditure to measure the firm’s

growth options. Specifically, GrowthExp equals the sum of the company’s R&D

(Compustat item 46) expenditure and capital expenditure (Compustat item 30) scaled by

the company’s total assets (Compustat annual item 6) of the most recent fiscal year. That

is, GrowthExp measures how much the company invests for the future. We expect

GrowthExp to be positively associated with the issuance of LTG forecasts.

We also include three control variables relating to a company’s growth options.

18

Hitech is a dummy variable that equals one for firms with Compustat SIC code

3570-3577 (computer hardware), or 7371-7379 (computer software), or 2833-2836

(pharmaceutical), and zero otherwise. B/M is the ratio of the company’s book value to

market value at the end of the most recent fiscal year. We obtain a company’s book value

(Compustat item 60) and market value (Compustat annual item 199*25) from the

Compustat database. Log(size) is the natural log of market value of equity (Compustat

annual item 199*25) in millions of dollars for the most recent fiscal year.

(f) Institutional Ownership

We collect the institution ownership information from the Thomson Financial

Ownership database. Institution equals the total number of shares held by institutions

who report their equity ownership in the quarterly 13f filings to the SEC divided by the

total number of shares outstanding at the end of the previous calendar year. For firms

with the institutional investor holdings data missing, we assume that these firms are

100% individually-owned and set Institution to zero.6

3.3 Summary statistics

To be included in our sample, an observation needs to have all the above-mentioned

variables available. We also delete 2,417 observations with negative book value and 69

observations with institutional holdings available but number of shares outstanding

missing. Our final sample includes 170,139 one-year-ahead analyst-firm-year

6 Ljungqvist et al (2005) suggest that it is possible that these companies are randomly missing. As a

robustness check, we delete observations with missing institutional ownership and our results are similar.

19

combinations.

Table 2 presents summary statistics. For the combined sample, 30.7 percent of the

firm-analyst-year combinations are associated with LTG forecasts. On average, the

analysts have issued forecasts for any company for approximately seven and a half years,

and issued forecasts for a particular company for more than four years. 35.2 percent of

the sample is associated with analysts hired by brokers who appear as top 15 in “the

leader list” of the Institutional Investor magazine (II) from year 1990 to year 2002. The

net forecast error of the most recent one-year-ahead forecasts the previous year is 67

cents for a stock priced at 100 dollars. The mean past forecast bias is negative, indicating

that the forecasts are optimistic, but the median is positive. On average, R&D and capital

expenditures account for 10.1 percent of total assets. 13.8 percent of sample is associated

with high technology companies. The mean percentage of institutional ownership is 52.6

percent.

4. Why do analysts issue LTG forecasts?

4.1 Univariate tests

We first conduct a series of univariate tests and report our results in Table 3. We

find that high-status broker affiliated analysts with more experience who issue more

accurate near-term forecasts in the past for the company are more likely to issue LTG

forecasts. We also find that analysts who are less optimistic about the company are more

20

likely to issue LTG forecasts. In addition, IB is significantly higher for the group with

LTG forecasts. Firms with more growth options (only median) and more stocks held by

institutional investors are more likely to receive LTG forecasts.

Overall, our univariate results largely support the analyst ability signaling,

management pleasing, and investor informational need satisfying hypotheses, but

contradict the analyst optimism revealing hypothesis.

4.2 Multivariate tests

We expect LTG issuance decisions to be partly driven by analyst peculiarities such

as their working habits or tastes, and thus focus on the controlling of analyst-level

heterogeneities. We estimate a fixed-effect model with analyst-year effect fixed.7 That is,

we focus on analysts’ decision to issue long-term forecasts among all the companies they

cover in a given year. As a robustness check, we re-estimate a fixed-effect and a random

effect model with only analyst effect, which allow us to include independent variables

that are within analyst-year groups such as Exp1 and Top15. To account for yearly

variations, we also include year dummies.

In column 1 of Table 4, we report the estimation results with analyst-year effect

fixed. 16,197 analyst-year pairs (80,224 observations) are dropped due to all positive or

all negative outcomes, but still 11,300 analyst-year pairs (89,915 observations) remain,

7 We also estimate a random-effect model including analyst effect as in Ljungqvist et al (2006). The results

are similar.

21

indicating that a given analyst may issue LTG forecasts for only a subset of companies

she covers in a given year. Therefore, the issuance decision of LTG forecasts goes beyond

analyst peculiarity.

Although LTG forecasts are documented as extremely inaccurate and overly

optimistic, analysts are more likely to choose the companies they had more accurate past

near-term forecasts for LTG coverage. However, analysts are less likely to issue LTG

forecasts as they gain more firm-specific experience for the company. This result may be

driven by analyst picking firms newly added to coverage for LTG forecasts.

We also find the estimated coefficient of Past_FB to be significantly negative,

indicating that analysts may be more likely to issue LTG forecasts for companies they are

more optimistic about.

We document strong support for the manager pleasing hypothesis. Investment

banking tie (IB) is significantly positive at the one percent level. The evidence regarding

the investor informational need satisfying hypothesis is, however, mixed. Analysts are

more likely to pick companies with higher institutional ownership. However, companies

with larger growth expenditures are less likely chosen for LTG coverage after controlling

for other firm characteristics such as size and B/M.

In Column 2 and 3, we report the estimation results from a fixed-effect model with

analyst effect fixed, and a random effect model including analyst effect. For both models,

we include year dummies, but do not report the estimated coefficients to conserve space.

Overall, the results are similar. We find support for the management pleasing and

22

optimism revealing motives, but mixed evidence regarding the analyst ability signaling

and investor informational needs satisfying motives. For example, we find that analysts

who have more general experience (only according to the random-effect model), who are

able to issue more accurate near-term forecasts in the past, and who are affiliated with

high status brokers are more likely to issue LTG forecasts, but again analysts seem to

drop LTG coverage as they gain more firm-specific experience. Regarding the investor

information needs satisfying hypothesis, we find that the coefficient of Institution is

significantly positive as expected, but the coefficient of Growth_Exp is insignificant.

Taken together, we find evidence for the manger pleasing and analyst optimism

revealing motives, but mixed evidence for investor informational needs satisfying and

analyst ability signaling motives.

4.3 Bubble period evidence

It is likely that analyst motives change depending upon market factors such as the

competitiveness in the underwriting market and the power of institutional investors.

Therefore, analysts may have extra incentives to please managers during the bubble

period. However, providing optimistic LTG forecasts is an implicit form of pleasing, and

analysts may go to the extreme of providing optimistic recommendations when they are

under extra pressure in the late nineties. Therefore, it is eventually an empirical question

whether analysts are more likely to provide LTG forecasts to please managers during the

bubble period. We introduce the dummy variables, Bubble, and its interactive terms with

23

IB. Following Bradley, Jordan, and Ritter (2006), we define the bubble period as year

1999 and 2000. Table 5 contains our results. We find no evidence indicating that LTG

forecasts are more motivated by the manager pleasing incentives during the bubble

period.

5. Institutional investors’ role in analysts’ motives to issue LTG forecasts

We introduce two explanatory variables: the interactive term between Institution

and GrowthExp, and the interactive term between Institution and IB. We expect the

estimated coefficient of Institution*GrowthExp to be positive and the estimated

coefficient of Institution*IB to be negative.

In Table 6, we find that companies with higher institutional ownership are less

likely to be chosen for LTG forecast coverage because of investment banking ties. In

addition, we show that institutional investors’ role goes beyond that. The coefficient of

the interactive term between institutional ownership and growth expenditure is

significantly positive, indicating that analysts are more likely to issue LTG forecasts for

companies with higher R&D and capital expenditures given the presence of higher

institutional ownership.

To summarize, our results confirm the important role institutional investors play in

analyst research. We find that institutional ownership is positively associated with LTG

issuance for the right reason (investor informational needs satisfying), but negatively

24

associated with LTG issuance for the wrong reason (manager pleasing).

6. Conclusion

This paper examines analysts’ motives to issue LTG forecasts. We develop four

non-exclusive hypotheses, which are that analysts issue early forecasts to signal their

ability, to reveal their optimism, to please the management (since these forecasts are

overly optimistic), and to satisfy investors’ informational needs. With one-year-ahead

annual earnings forecasts as our benchmark sample, we test our hypotheses using a

fixed-effect logit model with the analyst-year effect fixed, which ensures that our results

are not driven by analyst peculiarities such as their working habits that equally affect

analysts’ decision to issue long-term forecasts for all the companies they cover.

We find support for the manager pleasing and analyst optimism revealing

hypothesis, but mixed results for the ability signaling and investor informational needs

satisfying motives. In addition, we examine institutional investors’ role in determining

analysts’ motives to issue long-term forecasts. We find that analysts are less (more) likely

to issue long-term forecasts to companies with large institutional ownership to please

managers (to meet investors’ information needs).

This paper contributes to the literature in several ways. First, an examination of

the providence of long-term forecasts offers several advantages in investigating conflicts

of interests, and we show that long-term forecasts may serve as a manipulative tool for

25

analysts to please managers. In addition, our results augment Ljungqvist et al (2006)’s

finding about the role of institutional investors in analyst research.

Table 1. The Distribution of Long-term Forecasts by Calendar Year

Panel A, B, and C present the distribution of analyst-firm pairs that are associated with LTG

forecasts, analysts who issue LTG forecasts, and firms who receive LTG forecasts by calendar

year, respectively. We collect the one-year-ahead annual earnings forecasts (FY1) in the I/B/E/S

detail history file from year 1991 to 2003. We identify each analyst-firm-year combination and

check whether there are long horizon earnings growth forecasts (LTG), as reported in I/B/E/S,

associated with these analyst-firm-year combinations.

Analyst-firm pairs Analysts Firms

FY1 LTG Proportion

(%) FY1 LTG

Proportion

(%) FY1 LTG

Proportion

(%)

(1) (2) (3)=(2)/(1) (4) (5) (6)=(5)/(4) (7) (8) (9)=(8)/(7)

1991 7572 3278 43.29 350 189 54.00 480 393 81.88

1992 6940 3072 44.27 287 181 63.07 651 551 84.64

1993 10546 4394 41.67 431 250 58.00 663 535 80.69

1994 11366 4930 43.37 536 335 62.50 795 650 81.76

1995 13109 5498 41.94 600 364 60.67 928 684 73.71

1996 14567 6730 46.20 795 479 60.25 1163 867 74.55

1997 15312 7207 47.07 826 497 60.17 1057 705 66.70

1998 15482 6579 42.49 971 527 54.27 952 605 63.55

1999 15086 6686 44.32 947 531 56.07 692 500 72.25

2000 14985 6359 42.44 1081 648 59.94 686 471 68.66

2001 13274 6243 47.03 1132 684 60.42 280 206 73.57

2002 14331 6486 45.26 1575 926 58.79 329 220 66.87

2003 13285 4714 35.48 1758 879 50.00 391 166 42.46

Mean 12758 5552 43 868 499 58 697 504 72

Median 13285 6243 43 826 497 60 686 535 74

27

Table 2. Summary Statistics

Table 2 reports the summary statistics of our sample, which includes 170,139

analyst-firm-year observations over the period 1991-2003. LTG is a dummy variable that equals

one if the observation is associated with long-term earnings growth forecasts (LTG) as reported in

I/B/E/S, and zero otherwise. The general experience of the analysts (Exp1) is defined as the

number of years the analysts have issued earnings forecasts of any type for any company since

1983, when the sample period of I/B/E/S starts. Analysts’ firm-specific experience (Exp2) equals

the number of years the analysts have issued earnings forecasts of any type for the company since

1983. We define net forecast error (NFE) as 100 times the absolute value of the difference

between the actual earnings and the analyst forecasts divided by the company’s stock price the

company’s stock price at the end of the previous fiscal year. Past_NFE equals NFE t-1, that is, the

net forecast error of the most recent near-term earnings forecasts made during the previous year.

Forecast Bias (FB) is 100 times the difference between the actual earnings and the analyst

forecasts divided by the company’s stock price the company’s stock price at the end of the

previous fiscal year. We define FB t-1 as the past near-term forecast accuracy (Past_FB). We

define IB as a dummy variable that equals one if the analyst is affiliated with the investment bank

that serves as a book runner for the company’s new common stock issues, and zero otherwise.

IPO equals one for company i in year t if the company has an initial public offering as indicated

by the IPO flag in SDC for year t and t-1, and zero otherwise. Hitech is a dummy variable that

equals one for firms with Compustat SIC code 3570-3577 (computer hardware), or 7371-7379

(computer software), or 2833-2836 (pharmaceutical), and zero otherwise. B/M is the ratio of the

company’s book value to market value at the end of the most recent fiscal year. We obtain a

company’s book value (Compustat item 60) and market value (Compustat annual item 199*25)

from the Compustat database. GrowthExp equals the sum of the company’s R&D (Compustat

item 46) expenditure and capital expenditure (Compustat item 30) scaled by the company’s total

assets (Compustat annual item 6) of the most recent fiscal year. Log(size) is the natural log of

market value of equity (Compustat annual item 199*25) in millions of dollars of the most recent

fiscal year. Institution equals the total number of shares held by institutions who report their

equity ownership in quarterly 13f filings to the SEC divided by the total number of shares

outstanding at the end of the previous year. For firms with the institutional investors data missing,

we assume that these firms are 100% individually-owned and set Institution to zero.

Table 2 (Continue)

Variable Mean Std. Dev. 10% 25% Median 75% 90%

LTG 0.307 0.461 0 0 0 1 1

Exp1 7.46 4.67 2 4 7 11 14

Exp2 4.06 3.29 1 2 3 5 9

Top15 0.352 0.475 0 0 0 1 1

Past_nfe 0.667 3.322 0.008 0.054 0.164 0.485 1.320

Past_fb -0.082 3.387 -0.625 -0.099 0.036 0.213 0.643

IB 0.009 0.097 0 0 0 0 0

IPO 0.001 0.027 0 0 0 0 0

GrowthExp 0.101 0.010 0 0.032 0.078 0.143 0.220

Hitech 0.138 0.345 0 0 0 0 1

Log(size) 7.407 1.831 5.033 6.127 7.383 8.645 9.794

B/M 22.996 2395.12 0.142 0.253 0.424 0.642 0.909

Institution 0.526 0.227 0.210 0.380 0.551 0.687 0.793

Sample size 170139

Table 3. Why Do Analysts Issue LTG Forecasts? Univariate tests

Table 3 presents the results from a series of univariate tests. We report the mean and median value

for each subsample. Columns labeled as “Dif.” contain the difference of mean (medain) between

two subsamples. We report the t-statistics for means and an approximate z-statistic for a sum of

ranks test under the hypothesis that the distributions are equal. LTG is a dummy variable that

equals one if the observation is associated with long-term earnings growth forecasts (LTG) as

reported in I/B/E/S, and zero otherwise. The general experience of the analysts (Exp1) is defined

as the number of years the analysts have issued earnings forecasts of any type for any company

since 1983, when the sample period of I/B/E/S starts. Analysts’ firm-specific experience (Exp2)

equals the number of years the analysts have issued earnings forecasts of any type for the

company since 1983. We define net forecast error (NFE) as 100 times the absolute value of the

difference between the actual earnings and the analyst forecasts divided by the company’s stock

price the company’s stock price at the end of the previous fiscal year. Past_NFE equals NFE t-1,

that is, the net forecast error of the most recent near-term earnings forecasts made during the

previous year. The dummy variable Top15 takes on value one for analysts affiliated with the high

status brokers who appear as top 15 in “the leader list” of the Institutional Investor magazine (II),

and zero otherwise Forecast Bias (FB) is 100 times the difference between the actual earnings and

the analyst forecasts divided by the company’s stock price the company’s stock price at the end of

the previous fiscal year. We define FB t-1 as the past near-term forecast accuracy (Past_FB). IB is

defined as a dummy variable that equals one if the analyst is affiliated with the investment bank

that serves as a book runner for the company’s new common stock issues, and zero otherwise.

IPO equals one for company i in year t if the company has an initial public offering as indicated

by the IPO flag in SDC for year t and t-1, and zero otherwise. Hitech is a dummy variable that

equals one for firms with Compustat SIC code 3570-3577 (computer hardware), or 7371-7379

(computer software), or 2833-2836 (pharmaceutical), and zero otherwise. B/M is the ratio of the

company’s book value to market value at the end of the most recent fiscal year. We obtain a

company’s book value (Compustat item 60) and market value (Compustat annual item 199*25)

from the Compustat database. GrowthExp equals the sum of the company’s R&D (Compustat

item 46) expenditure and capital expenditure (Compustat item 30) scaled by the company’s total

assets (Compustat annual item 6) of the most recent fiscal year. Log(size) is the natural log of

market value of equity (Compustat annual item 199*25) in millions of dollars of the most recent

fiscal year. Institution equals the total number of shares held by institutions who report their

equity ownership in quarterly 13f filings to the SEC divided by the total number of shares

outstanding at the end of the previous year. For firms with the institutional investors data missing,

we assume that these firms are 100% individually-owned and set Institution to zero.

Table 3 (Continue)

Variable LTG=0 LTG=1 Dif T Dif Z

Mean

(1)

Median

(2)

Mean

(3)

Median

(4) (1)-(3) (2)-(4)

LTG 0.000 0 1.000 1

Exp1 7.389 7 7.618 7 -0.229 -9.34 0 -6.10

Exp2 4.048 3 4.088 3 -0.04 -2.34 0 1.05

Top15 0.328 0 0.406 0 -0.078 -31.55 0 -31.46

Past_nfe 0.744 0.185 0.492 0.127 0.252 14.44 0.058 43.21

Past_fb -0.101 0.036 -0.039 0.034 -0.062 -3.49 0.002 -1.83

IB 0.008 0 0.012 0 -0.004 -8.05 0 -8.05

IPO 0.001 0 0.001 0 0 -0.64 0 -0.64

GrowthExp 0.101 0.077 0.101 0.081 0 0.25 -0.004 -9.24

Hitech 0.130 0 0.157 0 -0.027 -15.17 0 -15.16

Bm 22.168 0.443 24.863 0.382 -2.695 -0.21 0.061 39.41

Logsize 7.305 7.285 7.635 7.610 -0.33 -34.42 -0.325 -33.11

Institution 0.517 0.544 0.546 0.567 -0.029 -23.81 -0.023 -22.81

Sample size 117882 52257

Table 4. Why Do Analysts Issue LTG Forecasts? Multivariate Tests

Table 4 present our results with LTG as dependent variable estimated from the fixed-effect model

with analyst-year effect fixed (Column 1), the fixed-effect model with analyst effect fixed

including yearly dummies (Column 2), and the random effect model including analyst effect with

yearly dummies (Column 3). We omit the estimated coefficients for the yearly dummies in

Column 2 and 3. LTG is a dummy variable that equals one if the observation is associated with a

long-term earnings growth forecast, and zero otherwise. The general experience of the analysts

(Exp1) is defined as the number of years the analysts have issued earnings forecasts of any type

for any company since 1983, when the sample period of I/B/E/S starts. Analysts’ firm-specific

experience (Exp2) equals the number of years the analysts have issued earnings forecasts of any

type for the company since 1983. We define net forecast error (NFE) as 100 times the absolute

value of the difference between the actual earnings and the analyst forecasts divided by the

company’s stock price the company’s stock price at the end of the previous fiscal year. Past_NFE

equals NFE t-1, that is, the net forecast error of the most recent near-term earnings forecasts made

during the previous year. The dummy variable Top15 takes on value one for analysts affiliated

with the high status brokers who appear as top 15 in “the leader list” of the Institutional Investor

magazine (II), and zero otherwise Forecast Bias (FB) is 100 times the difference between the

actual earnings and the analyst forecasts divided by the company’s stock price the company’s

stock price at the end of the previous fiscal year. We define FB t-1 as the past near-term forecast

accuracy (Past_FB). IB is defined as a dummy variable that equals one if the analyst is affiliated

with the investment bank that serves as a book runner for the company’s new common stock

issues, and zero otherwise. IPO equals one for company i in year t if the company has an initial

public offering as indicated by the IPO flag in SDC for year t and t-1, and zero otherwise. Hitech

is a dummy variable that equals one for firms with Compustat SIC code 3570-3577 (computer

hardware), or 7371-7379 (computer software), or 2833-2836 (pharmaceutical), and zero

otherwise. B/M is the ratio of the company’s book value to market value at the end of the most

recent fiscal year. We obtain a company’s book value (Compustat item 60) and market value

(Compustat annual item 199*25) from the Compustat database. GrowthExp equals the sum of the

company’s R&D (Compustat item 46) expenditure and capital expenditure (Compustat item 30)

scaled by the company’s total assets (Compustat annual item 6) of the most recent fiscal year.

Log(size) is the natural log of market value of equity (Compustat annual item 199*25) in millions

of dollars of the most recent fiscal year. Institution equals the total number of shares held by

institutions who report their equity ownership in quarterly 13f filings to the SEC divided by the

total number of shares outstanding at the end of the previous year. For firms with the institutional

investors data missing, we assume that these firms are 100% individually-owned and set

Institution to zero. For each model, we report the estimated coefficient, the z statistics, the

log-likelihood, and the sample size.

1 2 3

Predicted

Sign Coef. Z Coef. Z Coef. Z

Exp1 + -0.059 -1.44 0.005 1.95

Exp2 + -0.009 -2.86 -0.011 -4.24 -0.012 -4.44

Top15 + 0.076 2.77 0.114 5.39

Past_nfe - -0.024 -4.83 -0.022 -5.48 -0.028 -6.88

Past_fb - -0.008 -1.65 -0.012 -3.14 -0.014 -3.62

IB + 0.376 5.23 0.318 5.29 0.333 5.55

IPO + -0.098 -0.34 0.134 0.59 0.185 0.83

Hitech + 0.053 1.48 0.053 1.82 0.158 6.12

GrowthExp + -0.377 -3.47 -0.107 -1.23 -0.055 -0.65

Bm - 0.000 2.18 0.000 2.26 0.000 2.09

Logsize + 0.136 23.82 0.104 22.95 0.096 22.13

Institution + 0.276 6.86 0.217 6.66 0.281 8.88

-2.160 -41.56

Model

Analyst-year

Fixed effect

Analyst fixed effect

(with year dummies)

Analyst random effect

(with year dummies)

Log

-likelihood -37060 -70519 -86610

# of obs. 89915 140689 170139

33

Table 5. Bubble Period Evidence

We test whether managers have stronger incentive to issue LTG forecasts to please managers

during the bubble period by adding a dummy variable Bubble, which equals one for year 1999

and 2000 and zero otherwise, and its interactive term with IB. LTG is a dummy variable that

equals one if the observation is associated with a long-term earnings growth forecast, and zero

otherwise. The general experience of the analysts (Exp1) is defined as the number of years the

analysts have issued earnings forecasts of any type for any company since 1983, when the sample

period of I/B/E/S starts. Analysts’ firm-specific experience (Exp2) equals the number of years the

analysts have issued earnings forecasts of any type for the company since 1983. We define net

forecast error (NFE) as 100 times the absolute value of the difference between the actual earnings

and the analyst forecasts divided by the company’s stock price the company’s stock price at the

end of the previous fiscal year. Past_NFE equals NFE t-1, that is, the net forecast error of the most

recent near-term earnings forecasts made during the previous year. The dummy variable Top15

takes on value one for analysts affiliated with the high status brokers who appear as top

15 in “the leader list” of the Institutional Investor magazine (II), and zero otherwise

Forecast Bias (FB) is 100 times the difference between the actual earnings and the analyst

forecasts divided by the company’s stock price the company’s stock price at the end of the

previous fiscal year. We define FB t-1 as the past near-term forecast accuracy (Past_FB). IB is

defined as a dummy variable that equals one if the analyst is affiliated with the investment bank

that serves as a book runner for the company’s new common stock issues, and zero otherwise.

IPO equals one for company i in year t if the company has an initial public offering as indicated

by the IPO flag in SDC for year t and t-1, and zero otherwise. Hitech is a dummy variable that

equals one for firms with Compustat SIC code 3570-3577 (computer hardware), or 7371-7379

(computer software), or 2833-2836 (pharmaceutical), and zero otherwise. B/M is the ratio of the

company’s book value to market value at the end of the most recent fiscal year. We obtain a

company’s book value (Compustat item 60) and market value (Compustat annual item 199*25)

from the Compustat database. GrowthExp equals the sum of the company’s R&D

(Compustat item 46) expenditure and capital expenditure (Compustat item 30) scaled by

the company’s total assets (Compustat annual item 6) of the most recent fiscal year.

Log(size) is the natural log of market value of equity (Compustat annual item 199*25) in millions

of dollars of the most recent fiscal year. Institution equals the total number of shares held by

institutions who report their equity ownership in quarterly 13f filings to the SEC divided

by the total number of shares outstanding at the end of the previous year. For firms with

the institutional investors data missing, we assume that these firms are 100%

individually-owned and set Institution to zero. Bubble is a dummy variable that equals

one for year 1999 and 2000, zero otherwise. For each model, we report the estimated

coefficient, the z statistics, the log-likelihood, and the sample size.

1 2 3

Predicted Coef. Z Coef. Z Coef. Z

34

Sign

Exp1 + -0.059 -1.44 0.005 1.95

Exp2 + -0.009 -2.86 -0.011 -4.23 -0.012 -4.44

Top15 + 0.076 2.76 0.114 5.38

Past_nfe - -0.024 -4.82 -0.022 -5.47 -0.028 -6.87

Past_fb - -0.008 -1.65 -0.012 -3.13 -0.014 -3.62

IB + 0.402 5.13 0.351 5.37 0.360 5.52

IPO + -0.099 -0.35 0.133 0.59 0.185 0.83

Hitech + 0.053 1.48 0.053 1.81 0.158 6.13

GrowthExp + -0.377 -3.48 -0.107 -1.22 -0.054 -0.64

Bm - 0.000 2.18 0.000 2.26 0.000 2.09

Logsize + 0.136 23.82 0.104 22.96 0.096 22.13

Institution + 0.276 6.86 0.217 6.66 0.281 8.88

Bubble ? -0.313 -1.9 -0.102 -3.15

Bubble*IB + -0.165 -0.84 -0.208 -1.29 -0.174 -1.07

Constant -2.160 -41.56

Model

Analyst-year

Fixed effect

Analyst fixed

effect

(with year dummies)

Analyst random

effect

(with year

dummies)

Log

-likelihood -37059 -70523 -86616

# of obs. 89915 140689 170139

35

Table 6. The Role of Institutional Investors in Analysts’ Motive to Issue Long-term

Forecasts

We test the effect of institutional investors on analyst motives. LTG is a dummy variable that

equals one if the observation is associated with a long-term earnings growth forecast, and zero

otherwise. The general experience of the analysts (Exp1) is defined as the number of years the

analysts have issued earnings forecasts of any type for any company since 1983, when the sample

period of I/B/E/S starts. Analysts’ firm-specific experience (Exp2) equals the number of years the

analysts have issued earnings forecasts of any type for the company since 1983. We define net

forecast error (NFE) as 100 times the absolute value of the difference between the actual earnings

and the analyst forecasts divided by the company’s stock price the company’s stock price at the

end of the previous fiscal year. Past_NFE equals NFE t-1, that is, the net forecast error of the most

recent near-term earnings forecasts made during the previous year. The dummy variable Top15

takes on value one for analysts affiliated with the high status brokers who appear as top

15 in “the leader list” of the Institutional Investor magazine (II), and zero otherwise

Forecast Bias (FB) is 100 times the difference between the actual earnings and the analyst

forecasts divided by the company’s stock price the company’s stock price at the end of the

previous fiscal year. We define FB t-1 as the past near-term forecast accuracy (Past_FB). IB is

defined as a dummy variable that equals one if the analyst is affiliated with the investment bank

that serves as a book runner for the company’s new common stock issues, and zero otherwise.

IPO equals one for company i in year t if the company has an initial public offering as indicated

by the IPO flag in SDC for year t and t-1, and zero otherwise. Hitech is a dummy variable that

equals one for firms with Compustat SIC code 3570-3577 (computer hardware), or 7371-7379

(computer software), or 2833-2836 (pharmaceutical), and zero otherwise. B/M is the ratio of the

company’s book value to market value at the end of the most recent fiscal year. We obtain a

company’s book value (Compustat item 60) and market value (Compustat annual item 199*25)

from the Compustat database. GrowthExp equals the sum of the company’s R&D

(Compustat item 46) expenditure and capital expenditure (Compustat item 30) scaled by

the company’s total assets (Compustat annual item 6) of the most recent fiscal year.

Log(size) is the natural log of market value of equity (Compustat annual item 199*25) in millions

of dollars of the most recent fiscal year. Institution equals the total number of shares held by

institutions who report their equity ownership in quarterly 13f filings to the SEC divided

by the total number of shares outstanding at the end of the previous year. For firms with

the institutional investors data missing, we assume that these firms are 100%

individually-owned and set Institution to zero. For each model, we report the estimated

coefficient, the z statistics, the log-likelihood, and the sample size.

36

Predicted 1 2 3

Sign Coef. Z Coef. Z Coef. Z

Exp1 + -0.058 -1.43 0.005 1.94

Exp2 + -0.009 -2.86 -0.011 -4.25 -0.012 -4.45

Top15 + 0.076 2.77 0.114 5.3

Past_nfe - -0.024 -4.84 -0.022 -5.5 -0.028 -6.9

Past_fb - -0.008 -1.67 -0.012 -3.16 -0.014 -3.65

IB + 0.702 4.14 0.664 4.77 0.711 5.12

IPO + -0.072 -0.25 0.151 0.67 0.202 0.9

Hitech + 0.052 1.45 0.052 1.79 0.157 6.05

GrowthExp + -0.914 -4.41 -0.473 -2.83 -0.335 -2.08

Bm - 0.000 2.19 0.000 2.28 0.000 2.11

Logsize + 0.136 23.75 0.104 22.91 0.096 22.08

Institution + 0.171 3.13 0.148 3.35 0.230 5.33

Institution*IB - -0.633 -2.11 -0.688 -2.74 -0.757 -3.01

Institution*

GrowthExp + 1.128 3.08 0.765 2.59 0.596 2.07

constant -2.135 -39.27

Model

Analyst-year

Fixed effect

Analyst fixed effect(with

year dummies)

Analyst random effect

(with year dummies)

Log-

likelihood -37053 -70512 -86610

# of obs. 89915 140689 170139

37

References

Agrawal, Anup, and Mark A. Chen, 2005a. Analyst conflicts and research quality,

Working paper. The University of Alabama.

Agrawal, Anup, and Mark A. Chen, 2005b. Do analyst conflicts matter? evidence from

stock recommendations. Working paper. The University of Alabama.

Asquith, P., M. Mikhail, and A. Au, 2005. Information content of equity analyst reports.

Journal of Financial Economics 75, 245-282.

Bradley, Daniel J., Bradford D. Jordan, and Jay R. Ritter, 2006. Analyst behavior

following IPOs: the "bubble period" evidence. forthcoming, Review of Financial

Studies

Chan, Louis K. C., Jason Karceski, and Josef Lakonishok, 2003. The level and

persistence of growth rates. Journal of Finance 58, 643-684.

Clement, Michael B., 1999. Analyst forecast accuracy: do ability, resources, and portfolio

complexity matter? Journal of Accounting and Economics 27, 285-303.

Davis, Gerald F. and E. Han Kim, 2006, Business ties and proxy voting by mutual funds.

forthcoming, Journal of Financial Economics.

Dechow, Patricia M., and Richard G. Sloan, 1997. Returns to contrarian investment

strategies: tests of naive expectations hypotheses. Journal of Financial Economics

43, 3-27.

Defond, Mark L., and Mingyi Hung, 2003. An empirical analysis of analysts' cash flow

forecasts. Journal of Accounting and Economics 35, 73-100.

Dugar, Amitabh, and Siva Nathan, 1995, The effect of investment banking relationships

on financial analysts' earnings forecasts and investment recommendations.

Contemporary Accounting Research 12, 131-160.

Harris, Richard D. F., 1999. The accuracy, bias and efficiency of analysts’ long run

earnings growth forecasts. Journal of Business Finance & Accounting 26, 725-755

James, Christopher M., and Jason J. Karceski, 2006. Strength of analyst coverage

following IPOs. Journal of Financial Economics 82, 1-34.

Kang, Sok Hyon, John O'Brien, and K. Sivaramakrishnan, 1994. Analysts' interim

earnings forecasts: evidence on the forecasting process. Journal of Accounting

Research 32, 103-112.

La Porta, Rafael, 1996. Expectations and the cross-section of stock returns. Journal of

38

Finance 51, 1715-1742.

Lim, Terence, 2001. Rationality and analysts' forecast bias. Journal of Finance 56,

369-385.

Lin, Hsiou Wei, and Maureen F. McNichols, 1998. Underwriting relationships, analysts'

earnings forecasts and investment recommendations. Journal of Accounting and

Economics 25, 101-127.

Ljungqvist, Alexander, Marston, Felicia C., Starks, Laura T., Wei, Kelsey D. and Yan,

Hong. Conflicts of interest in sell-side research and the moderating role of

institutional investors. Forthcoming, Journal of Financial Economics

McNichols, Maureen, and Patricia C. O'Brien, 1997. Self-selection and analyst coverage.

Journal of Accounting Research 35, 167-199.

Michaely, Roni, and Kent L. Womack, 1999. Conflict of interest and the credibility of

underwriter analyst recommendations. Review of Financial Studies 12, 653-686.

Mikhail, Michael B., Beverly R. Walther, and Richard H. Willis, 2003. The effect of

experience on security analyst underreaction. Journal of Accounting and Economics

35, 101-116.

Mikhail, Michael B., Beverly R. Walther, and Richard H. Willis, 2004. Do security

analysts exhibit persistent differences in stock picking ability? Journal of Financial

Economics 74, 67-91.

Mikhail, Michael B., Beverly R. Walther, and Richard H. Willis, 2006. When security

analysts talk who listens? Working paper, Arizona State University.

O'Brien, Patricia C., Maureen F. McNichols, and Hsiou Wei Lin, 2005. Analyst

impartiality and investment banking relationships. Journal of Accounting Research

43, 623-650.

Purnanandam, Amiyatosh K., and Bhaskaran Swaminathan, 2004. Are IPOs really

underpriced? Review of Financial Studies 17, 811-848.

Raedy, Jana S., Philip B. Shane, and Yanhua Yang, 2006. Horizon-dependent

underreaction in financial analysts' earnings forecasts. Contemporary Accounting

Research 23, 291-322.

Rajan, Raghuram, and Henri Servaes, 1997. Analyst following of initial public offerings.

Journal of Finance 52, 507-529.

Richardson, Scott, Siew Hong Teoh, and Peter D. Wysocki, 2004. The walk down to

beatable analyst forecasts: the role of equity issuance and insider trading incentives.

39

Contemporary Accounting Research 21, 885-924.

Sinha, P., L. D. Brown, and D Das, 1997. A re-examination of financial analysts'

differential earnings forecast accuracy. Contemporary Accounting Research 14,

1-42.